Land subtraction runtime cleanup and exploration drivers#20
Merged
Conversation
- close A-E run-report defects and subagent review findings - collapse final display/projection to a single answer authority - keep verification/source/latent signals advisory for answer-first exploration - update tests, public hygiene, and source-current sync
- allow explicit seed policies such as 24 batches to bypass the old 16-batch clamp - keep an upper safety bound and add regression coverage - record the fresh-run discovery in the debt ledger as closed
…tatus failure - nextgen.py: bump CANONICAL_FAMILY_VERSION v1->v2 so candidates carrying the old 6-bucket canonical recompute against the current 8-bucket mechanism (d20b959 widened buckets 6->8 without a version bump, leaving resume/replay mixing buckets) - engine_runner.py: redact() the non-streaming api-request.json dump to match the streaming path (previously wrote raw prompt/request to disk unredacted) - jobs.py: log job-status persistence failures via logger.warning(exc_info) instead of silently swallowing them; best-effort write stays non-fatal Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…e bindings + QD applied bin P0 #2: move cbt_soft_budget_adjustment out of reproductive_value into an ordering-only bonus in ParentSelector.select, so the canonical search-diversity soft quota reorders parents but never gates the >=0 parent-eligibility floor. P0 #3: thread the real project root into source-binding annotation via a runtime-only ArchiveManager.project_root (excluded from to_dict/from_dict), and drop the no-op annotate side effect from the parent selector. P1 #7: keep the applied/not_applied token in candidate_bin_key ([1:5]) so applied state splits the QD bin; bound the pool-preprocess payload when many bins inflate the coverage report; add a regression test. P2 #9: mark the model-facing coverage_gate as advisory_pressure_only instead of renaming the protocol key (no model-protocol break). Mechanical fixes: API port env alignment (#5), VerifierStack->NexusRuntime naming (#8), order-independent novelty ratio + double-prefix metric key (#10), README quickstart env (#11), runtime task.yaml warning log (#12), pytest cacheprovider hardening (#13). Cross-model reviewed (codex) for the protected-file changes (#2/#3/#7/#9); canonical/cbt soft signals verified to never touch eligibility/stop/final gates. Full suite: 659 passed, 1 skipped. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
zaler23
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Land the CognitiveEvolve subtraction/runtime cleanup and exploration-driver work on
mzz/seed-coverage-resurrection-gate.Major scope:
source_contextthrough seed, mutation planning, and offspring prompts.Validation
Local:
git diff --checkpassed.UV_PROJECT_ENVIRONMENT=/tmp/cogev-github-final-venv PYTHONDONTWRITEBYTECODE=1 uv run --extra test python -B -m pytest -q -p no:cacheprovider→659 passed, 1 skipped.UV_PROJECT_ENVIRONMENT=/tmp/cogev-github-final-venv uv run --extra test python -B scripts/cogev.py doctor --scope all→49/49 checks passed.origin/main...HEAD: no untracked files, no tracked.env, no high-risk secret/local-path hits.GitHub:
CLEAN/MERGEABLE.Safety notes